Exit policy backtest
The six ways of managing a trade that are simulated on every past signal of a pair, and how the winner reaches the live plan.
The question it answers
Given the same entries, is it better to hold for TP1 with a tight stop, to give the trade room, to scale out, or to trail? The answer differs per pair, so the engine tests it per pair on every historical confirmed-pattern signal and shows the winner on the live consensus card.
The six policies
| Policy | Rule |
|---|---|
| fixed stop (tight) and TP1 | Stop at the tight level, target TP1. This is the default plan. |
| wide stop and TP1 | Stop at the wide level, target TP1. |
| half out at TP1, rest trails 1.5 ATR with stop at break-even | Close half at TP1, move the stop to entry, let the rest trail 1.5 ATR behind the best price. |
| stop to break-even after +1 ATR, aim for TP2 | Once price is 1 ATR in favour the stop moves to entry; target TP2. |
| fixed, closed after 3x expected bars | As fixed, but closed at market after three times the pattern's median bars-to-target if nothing happened. |
| stop beyond the winners' typical adverse wobble (p80), TP1 | Stop at 1.1x the 80th-percentile adverse excursion of this pair's winning patterns (needs at least 20 winners), target TP1. |
Each simulation uses the pair's stored patterns on 15m, 1h and 4h, the real trade plan built at the completion bar, entry only inside the entry range within the signal expiry, the taker fee, slippage scaled by session (Asia 1.3x, Europe 1.0x, US 0.8x, weekend 1.6x) and the funding cost of the holding time (average stored funding, else 0.01% per 8 hours).
Where to see it
Performance > Exit policies. Choose the pair in the selector. The first line states The live plan for this pair uses: … with the number of signals, the fee, the funding assumption and when it was computed. The table lists per policy:
- n: simulated trades;
- win: share closed with a profit;
- expectancy: average return per trade, unleveraged, after costs;
- avg win, avg loss;
- PF: profit factor (gross profit over gross loss);
- med bars: median bars held.
The best badge marks the policy with the highest expectancy among those with at least 30 trades. Policies with fewer than 30 trades are shown but not eligible.
Re-run simulates all six policies again on the pair's current stored patterns, useful after a deep analysis has added history. The test also runs automatically during deep analysis (stage backtesting exit policies).
run deep analysis first means the pair has no stored patterns yet.
How it reaches the live plan
The winner is stored per pair and shown on the consensus card as Exit: … · won n% · expectancy +x% (n). The card's ranges are unchanged; the line tells you how to manage the trade after entry. Until a policy reaches 30 trades the card does not show the line and the default fixed policy applies.
Reading the results honestly
- A wide stop that wins more often but with a worse expectancy is not better. Expectancy after costs is the ranking measure.
- partial_trail often has the best expectancy on trending pairs and the worst on ranging ones; the breakdown by pair on the backtest page tells you which you have.
- The wobble_stop result is a good sanity check on the tight stop. If it beats fixed by a wide margin, the tight stop is being hit by noise, and the analog widening on the consensus card is doing useful work.
- Thirty trades is the minimum for eligibility, not a large sample. Prefer policies that stay best after a re-run with more history.
Note: the exit-policy test uses pattern signals. The consensus card's own stop and target ranges come from the median excursions after correct calls, so the two methods can disagree on how wide a stop should be; the card takes the wider of its own stop and the analog stop.